From 31439046e0948cf2ca13586a0c935f5f23b1f346 Mon Sep 17 00:00:00 2001 From: Andy Polyakov Date: Tue, 17 Oct 2006 14:37:07 +0000 Subject: [PATCH] bn/asm/ppc.pl to use ppc-xlate.pl. --- crypto/bn/asm/ppc.pl | 224 +++++++++++++------------------------------ 1 file changed, 64 insertions(+), 160 deletions(-) diff --git a/crypto/bn/asm/ppc.pl b/crypto/bn/asm/ppc.pl index 08e0053473..0c0206b0d7 100644 --- a/crypto/bn/asm/ppc.pl +++ b/crypto/bn/asm/ppc.pl @@ -151,91 +151,15 @@ if ($opf =~ /32\.s/) { $TR= "td"; # conditional trap } else { die "nonsense $opf"; } -( defined shift || open STDOUT,">$opf" ) || die "can't open $opf: $!"; +$0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1; +( $xlate="${dir}ppc-xlate.pl" and -f $xlate ) or +( $xlate="${dir}../../perlasm/ppc-xlate.pl" and -f $xlate) or +die "can't locate ppc-xlate.pl"; -# function entry points from the AIX code -# -# There are other, more elegant, ways to handle this. We (IBM) chose -# this approach as it plays well with scripts we run to 'namespace' -# OpenSSL .i.e. we add a prefix to all the public symbols so we can -# co-exist in the same process with other implementations of OpenSSL. -# 'cleverer' ways of doing these substitutions tend to hide data we -# need to be obvious. -# -my @items = ("bn_sqr_comba4", - "bn_sqr_comba8", - "bn_mul_comba4", - "bn_mul_comba8", - "bn_sub_words", - "bn_add_words", - "bn_div_words", - "bn_sqr_words", - "bn_mul_words", - "bn_mul_add_words"); +( defined shift || open STDOUT,"| $^X $xlate $opf" ) || + die "can't call $xlate: $!"; -if ($opf =~ /linux/) { do_linux(); } -elsif ($opf =~ /aix/) { do_aix(); } -elsif ($opf =~ /osx/) { do_osx(); } -else { do_bsd(); } - -sub do_linux { - $d=&data(); - - if ($BITS==64) { - foreach $t (@items) { - $d =~ s/\.$t:/\ -\t.section\t".opd","aw"\ -\t.align\t3\ -\t.globl\t$t\ -$t:\ -\t.quad\t.$t,.TOC.\@tocbase,0\ -\t.size\t$t,24\ -\t.previous\n\ -\t.type\t.$t,\@function\ -\t.globl\t.$t\ -.$t:/g; - } - } - else { - foreach $t (@items) { - $d=~s/\.$t/$t/g; - } - } - # hide internal labels to avoid pollution of name table... - $d=~s/Lppcasm_/.Lppcasm_/gm; - print $d; -} - -sub do_aix { - # AIX assembler is smart enough to please the linker without - # making us do something special... - print &data(); -} - -# MacOSX 32 bit -sub do_osx { - $d=&data(); - # Change the bn symbol prefix from '.' to '_' - foreach $t (@items) { - $d=~s/\.$t/_$t/g; - } - # Change .machine to something OS X asm will accept - $d=~s/\.machine.*/.text/g; - $d=~s/\#/;/g; # change comment from '#' to ';' - print $d; -} - -# BSD (Untested) -sub do_bsd { - $d=&data(); - foreach $t (@items) { - $d=~s/\.$t/_$t/g; - } - print $d; -} - -sub data { - local($data)=< 0 then result !=0 # In either case carry bit is set. - bc BO_IF,CR0_EQ,Lppcasm_sub_adios + beq Lppcasm_sub_adios addi r4,r4,-$BNSZ addi r3,r3,-$BNSZ addi r5,r5,-$BNSZ @@ -1635,11 +1546,11 @@ Lppcasm_sub_mainloop: # if carry = 1 this is r7-r8. Else it # is r7-r8 -1 as we need. $STU r6,$BNSZ(r3) - bc BO_dCTR_NZERO,CR0_EQ,Lppcasm_sub_mainloop + bdnz- Lppcasm_sub_mainloop Lppcasm_sub_adios: subfze r3,r0 # if carry bit is set then r3 = 0 else -1 andi. r3,r3,1 # keep only last bit. - bclr BO_ALWAYS,CR0_LT + blr .long 0x00000000 @@ -1670,7 +1581,7 @@ Lppcasm_sub_adios: # check for r6 = 0. Is this needed? # addic. r6,r6,0 #test r6 and clear carry bit. - bc BO_IF,CR0_EQ,Lppcasm_add_adios + beq Lppcasm_add_adios addi r4,r4,-$BNSZ addi r3,r3,-$BNSZ addi r5,r5,-$BNSZ @@ -1680,10 +1591,10 @@ Lppcasm_add_mainloop: $LDU r8,$BNSZ(r5) adde r8,r7,r8 $STU r8,$BNSZ(r3) - bc BO_dCTR_NZERO,CR0_EQ,Lppcasm_add_mainloop + bdnz- Lppcasm_add_mainloop Lppcasm_add_adios: addze r3,r0 #return carry bit. - bclr BO_ALWAYS,CR0_LT + blr .long 0x00000000 # @@ -1707,24 +1618,24 @@ Lppcasm_add_adios: # r5 = d $UCMPI 0,r5,0 # compare r5 and 0 - bc BO_IF_NOT,CR0_EQ,Lppcasm_div1 # proceed if d!=0 + bne Lppcasm_div1 # proceed if d!=0 li r3,-1 # d=0 return -1 - bclr BO_ALWAYS,CR0_LT + blr Lppcasm_div1: xor r0,r0,r0 #r0=0 li r8,$BITS $CNTLZ. r7,r5 #r7 = num leading 0s in d. - bc BO_IF,CR0_EQ,Lppcasm_div2 #proceed if no leading zeros + beq Lppcasm_div2 #proceed if no leading zeros subf r8,r7,r8 #r8 = BN_num_bits_word(d) $SHR. r9,r3,r8 #are there any bits above r8'th? $TR 16,r9,r0 #if there're, signal to dump core... Lppcasm_div2: $UCMP 0,r3,r5 #h>=d? - bc BO_IF,CR0_LT,Lppcasm_div3 #goto Lppcasm_div3 if not + blt Lppcasm_div3 #goto Lppcasm_div3 if not subf r3,r5,r3 #h-=d ; Lppcasm_div3: #r7 = BN_BITS2-i. so r7=i cmpi 0,0,r7,0 # is (i == 0)? - bc BO_IF,CR0_EQ,Lppcasm_div4 + beq Lppcasm_div4 $SHL r3,r3,r7 # h = (h<< i) $SHR r8,r4,r8 # r8 = (l >> BN_BITS2 -i) $SHL r5,r5,r7 # d<<=i @@ -1741,7 +1652,7 @@ Lppcasm_divouterloop: $SHRI r11,r4,`$BITS/2` #r11= (l&BN_MASK2h)>>BN_BITS4 # compute here for innerloop. $UCMP 0,r8,r9 # is (h>>BN_BITS4)==dh - bc BO_IF_NOT,CR0_EQ,Lppcasm_div5 # goto Lppcasm_div5 if not + bne Lppcasm_div5 # goto Lppcasm_div5 if not li r8,-1 $CLRU r8,r8,`$BITS/2` #q = BN_MASK2l @@ -1762,9 +1673,9 @@ Lppcasm_divinnerloop: # the following 2 instructions do that $SHLI r7,r10,`$BITS/2` # r7 = (t<>BN_BITS4) - $UCMP 1,r6,r7 # compare (tl <= r7) - bc BO_IF_NOT,CR0_EQ,Lppcasm_divinnerexit - bc BO_IF_NOT,CR1_FEX,Lppcasm_divinnerexit + $UCMP cr1,r6,r7 # compare (tl <= r7) + bne Lppcasm_divinnerexit + ble cr1,Lppcasm_divinnerexit addi r8,r8,-1 #q-- subf r12,r9,r12 #th -=dh $CLRU r10,r5,`$BITS/2` #r10=dl. t is no longer needed in loop. @@ -1773,14 +1684,14 @@ Lppcasm_divinnerloop: Lppcasm_divinnerexit: $SHRI r10,r6,`$BITS/2` #t=(tl>>BN_BITS4) $SHLI r11,r6,`$BITS/2` #tl=(tl<=tl) goto Lppcasm_div7 + bge cr1,Lppcasm_div7 # if (l>=tl) goto Lppcasm_div7 addi r12,r12,1 # th++ Lppcasm_div7: subf r11,r11,r4 #r11=l-tl - $UCMP 1,r3,r12 #compare h and th - bc BO_IF_NOT,CR1_FX,Lppcasm_div8 #if (h>=th) goto Lppcasm_div8 + $UCMP cr1,r3,r12 #compare h and th + bge cr1,Lppcasm_div8 #if (h>=th) goto Lppcasm_div8 addi r8,r8,-1 # q-- add r3,r5,r3 # h+=d Lppcasm_div8: @@ -1791,12 +1702,12 @@ Lppcasm_div8: # the following 2 instructions will do this. $INSR r11,r12,`$BITS/2`,`$BITS/2` # r11 is the value we want rotated $BITS/2. $ROTL r3,r11,`$BITS/2` # rotate by $BITS/2 and store in r3 - bc BO_dCTR_ZERO,CR0_EQ,Lppcasm_div9#if (count==0) break ; + bdz Lppcasm_div9 #if (count==0) break ; $SHLI r0,r8,`$BITS/2` #ret =q<> 2 - bc BO_IF,CR0_EQ,Lppcasm_mw_REM + beq Lppcasm_mw_REM mtctr r7 Lppcasm_mw_LOOP: #mul(rp[0],ap[0],w,c1); @@ -1896,11 +1807,11 @@ Lppcasm_mw_LOOP: addi r3,r3,`4*$BNSZ` addi r4,r4,`4*$BNSZ` - bc BO_dCTR_NZERO,CR0_EQ,Lppcasm_mw_LOOP + bdnz- Lppcasm_mw_LOOP Lppcasm_mw_REM: andi. r5,r5,0x3 - bc BO_IF,CR0_EQ,Lppcasm_mw_OVER + beq Lppcasm_mw_OVER #mul(rp[0],ap[0],w,c1); $LD r8,`0*$BNSZ`(r4) $UMULL r9,r6,r8 @@ -1912,7 +1823,7 @@ Lppcasm_mw_REM: addi r5,r5,-1 cmpli 0,0,r5,0 - bc BO_IF,CR0_EQ,Lppcasm_mw_OVER + beq Lppcasm_mw_OVER #mul(rp[1],ap[1],w,c1); @@ -1926,7 +1837,7 @@ Lppcasm_mw_REM: addi r5,r5,-1 cmpli 0,0,r5,0 - bc BO_IF,CR0_EQ,Lppcasm_mw_OVER + beq Lppcasm_mw_OVER #mul_add(rp[2],ap[2],w,c1); $LD r8,`2*$BNSZ`(r4) @@ -1939,7 +1850,7 @@ Lppcasm_mw_REM: Lppcasm_mw_OVER: addi r3,r12,0 - bclr BO_ALWAYS,CR0_LT + blr .long 0x00000000 # @@ -1964,7 +1875,7 @@ Lppcasm_mw_OVER: xor r0,r0,r0 #r0 = 0 xor r12,r12,r12 #r12 = 0 . used for carry rlwinm. r7,r5,30,2,31 # num >> 2 - bc BO_IF,CR0_EQ,Lppcasm_maw_leftover # if (num < 4) go LPPCASM_maw_leftover + beq Lppcasm_maw_leftover # if (num < 4) go LPPCASM_maw_leftover mtctr r7 Lppcasm_maw_mainloop: #mul_add(rp[0],ap[0],w,c1); @@ -2017,11 +1928,11 @@ Lppcasm_maw_mainloop: $ST r11,`3*$BNSZ`(r3) addi r3,r3,`4*$BNSZ` addi r4,r4,`4*$BNSZ` - bc BO_dCTR_NZERO,CR0_EQ,Lppcasm_maw_mainloop + bdnz- Lppcasm_maw_mainloop Lppcasm_maw_leftover: andi. r5,r5,0x3 - bc BO_IF,CR0_EQ,Lppcasm_maw_adios + beq Lppcasm_maw_adios addi r3,r3,-$BNSZ addi r4,r4,-$BNSZ #mul_add(rp[0],ap[0],w,c1); @@ -2036,7 +1947,7 @@ Lppcasm_maw_leftover: addze r12,r10 $ST r9,0(r3) - bc BO_dCTR_ZERO,CR0_EQ,Lppcasm_maw_adios + bdz Lppcasm_maw_adios #mul_add(rp[1],ap[1],w,c1); $LDU r8,$BNSZ(r4) $UMULL r9,r6,r8 @@ -2048,7 +1959,7 @@ Lppcasm_maw_leftover: addze r12,r10 $ST r9,0(r3) - bc BO_dCTR_ZERO,CR0_EQ,Lppcasm_maw_adios + bdz Lppcasm_maw_adios #mul_add(rp[2],ap[2],w,c1); $LDU r8,$BNSZ(r4) $UMULL r9,r6,r8 @@ -2062,17 +1973,10 @@ Lppcasm_maw_leftover: Lppcasm_maw_adios: addi r3,r12,0 - bclr BO_ALWAYS,CR0_LT + blr .long 0x00000000 .align 4 EOF - $data =~ s/\`([^\`]*)\`/eval $1/gem; - - # if some assembler chokes on some simplified mnemonic, - # this is the spot to fix it up, e.g.: - # GNU as doesn't seem to accept cmplw, 32-bit unsigned compare - $data =~ s/^(\s*)cmplw(\s+)([^,]+),(.*)/$1cmpl$2$3,0,$4/gm; - # assembler X doesn't accept li, load immediate value - #$data =~ s/^(\s*)li(\s+)([^,]+),(.*)/$1addi$2$3,0,$4/gm; - return($data); -} +$data =~ s/\`([^\`]*)\`/eval $1/gem; +print $data; +close STDOUT; -- 2.25.1